Originally written as memo project for Quant II at Columbia SIPA, as response to prompt to evaluate MFI research for a hypothetical World Bank project.
As the World Bank considers expanding its microfinance operations, we must first evaluate the evidence on whether microfinance actually has positive impacts on its recipients’ material well-being.
Three recent studies shed light on microfinance’s impacts as measured by wealth, income, and expenditures: Banerjee and Duflo’s experimental 2010 research in which 104 slums in Hyderabad, India, were randomly selected for the opening of a microfinance institution (MFI); and Kondo (2008) and Montgomery’s (2015) two quasi-experimental studies that attempt to remove bias from observational data using a difference-in-differences model.
Findings from the three studies are mixed. The strongest and most consistent impacts are on household expenditures related to starting or improving businesses, and related entrepreneurial activity. Banerjee and Duflo found that in MFI treatment areas, there was statistically significant evidence of a shift in expenditure composition, with increased spending on durables and reduced spending on “temptation goods” and festivals. Montgomery found that MFI “yields the most impact for urban households running microenterprises and for very poor borrowers engaged in agriculture.” And Kondo found a “very significant positive” impact on household business enterprises. All of these findings suggest that, for households with a propensity to invest in a business, microcredit may spark increased business expenditures with the possibility of leading to longer-term material improvements for both households and communities.
Results concerning broader impacts on income, wealth or expenditures are less consistent. Banerjee and Duflo found no significant impact of MFI on total expenditure amounts. Kondo found “a mildly significant positive impact on per capita income, per capita total expenditure and per capita food expenditure,” yet this impact was regressive: benefits were found exclusively among the richest quartile of households, and in fact poorer households saw negative effects. Montgomery’s Pakistan results - despite a rosy conclusion that MFI participation “has positive impacts on both economic and social indicators of welfare” - is also fairly weak. The only statistically significant impact on expenditures was that the study’s “core poor” group, or households in the bottom quintile of the population, increased educational expenditures - yet this was in comparison to average MFI participants who were found to have a statistically significant reduction in educational expenditures compared to non-participants!
Assessing these impacts is complicated by various limitations and flaws in the studies reviewed. Banerjee and Duflo look at a relatively short time period of 15-18 months from the introduction of MFIs to the final survey. This short study length would fail to reveal any long-term expansion of consumption, as theoretically microfinance recipients may reduce short-term expenditures in order to save for longer-term plans; and it would not reveal long-term negative impacts such as indebtedness arising from microfinance loans.
Kondo’s study has a potentially serious flaw. The sampling of non-participating households in both “treatment” and “control” villages was not drawn from a random pool of villagers, but rather from lists of households “identified by MFI field personnel, center or [village] leaders.” This raises the possibility that samples were drawn from a population intentionally skewed by MFI staff in order to overstate the effects of microfinance.
The difference-in-differences model used by Kondo and Montgomery also has potential design issues:
- The studies’ use of “treatment” and “control” villages assumes similar characteristics in MFI participants after controlling for village fixed effects – but people selected in the earlier stages of an MFI program may differ from later selectees in ways that are not controlled by village fixed effects. If MFIs relaxed their selection process, or people selected earlier are for some other reason more responsive to microcredit, this could lead to overstated findings.
- Being selected for an MFI program may affect people’s behavior in ways that overstate impacts. If someone selected (but not yet receiving) microfinance cuts back on spending to prepare for investing in a business, that could narrow differences in the control group and lead to overstated expenditure impacts.
All three studies must also be evaluated in terms of local contexts, and great care should be taken before applying findings to new settings. For example, Banerjee and Duflo’s study is from Hyderabad, the capital of the Indian state where microfinance has expanded the fastest. Not only are the people there likely familiar with the microcredit model, they are also entrepreneurial: As the study puts it, “31% of households ran at least one small business at the baseline, compared to an OECD-country average of 12%.” And the Kondo and Montgomery studies both depend on data from villages where MFI programs either already existed or were planning to move, meaning that impacts may not hold for villages not already selected for an MFI program.
In conclusion, these studies provide some helpful evidence for considering microfinance expansion. The strongest evidence from these studies is that microfinance can be an important tool for improving material well-being in places where people are predisposed to make business investments and where microfinance has an existing foothold. The fact that this finding emerges both from a well-designed short-term experimental study, as well as two longer-view quasi-experimental studies, is quite promising.
On the other hand, we should be skeptical that microfinance, by itself, can raise basic living standards. The three studies have weak and mixed results on microfinance’s ability to raise baseline consumption or expenditures, and there is some evidence of microfinance causing harm. Additionally, all three studies take place in contexts where microfinance already exists, and all depend on people self-selecting their membership in loan programs. There is no evidence presented here that creating a microfinance program in a randomly selected poor region, or pushing people to join an MFI, would have positive impacts.
Microfinance may be able to play a role in helping some households in specific circumstances, but this research does not suggest that microfinance can necessarily expand basic material wealth without strongly considering population and implementation.